My work considers questions such as: How do the abstract component gestures of a phonological segment vary in their dynamical control as a function of phrasal structure? How do the structured surface variations guide listener perception and play a potential role in sound change? How are acoustic and articulatory prosodic signatures shaped by the language-specific phonological system? How is accommodation between conversing individuals related to prosodically salient positions? I pursue this work in the domain of F0 and articulation through empirical experimentation using laboratory techniques including electromagnetic articulography (EMA), real-time magnetic resonance imaging (rtMRI), and response mouse-tracking methods. I place an importance on data-driven computational modeling to further a unified, coherent account of suprasegmental and word- and phrase-level speech production.
Dissertation: The Prosodic Substrate of Consonant and Tone Dynamics
In my dissertation [.pdf], I employed a dynamic approach to relating consonant and tone by examining the segmental and tonal sensitivity to phrasal prosodic structure in a language with specified (i.e., relatively rigid) intonational patterns. To tackle the intricacies of segment/prosody interaction, I look at intonation patterns in the contemporary Seoul dialect of Korean. Given the segmental tone (i.e., F0) contrast for LAX versus TENSE stops, the convolution of these stops with the language’s strict Accentual Phrase (AP) tonal pattern provides a test bed for examining the co-expression of segmental and prosodic tonal specifications. In addition to clarifying the phonetic description needed for new pronunciation norms of younger generations speakers of this language, this work establishes how the local phonetic organization of a system of contrast is modulated as a function of phrasal positions by uncovering the tone gestures deployed for segmental "tenseness" and how these interact with the language's AP intonational patterning. Taken together, these results uncover an intricate tone pattern in which local segmental information is co-expressed with phrasal information and shows that dynamical modeling has the ability to account for how both categorical and gradient aspects of tone realization and how a prosodically asymmetric pattern emerges from a single underlying system. (Associated Work: Lee, Goldstein & Byrd, in preparation, talk at the LSA 2019, talk at the LSA 2020; Oh & Lee, 2018, 2020)
Prosodic Structuring in Spoken Language
In empirical work separate from the dissertation, I have examined other aspects of prosodic structuring in spoken language.
In my work on phonetic signatures of prosodic boundary and accentual prominence in Korean and English (Lee, 2011, 2013; Cho, Lee & Kim, 2011, 2014), I show that prosodically structured variability can be accounted for by the coordination between prosodic gestures and vocal tract gestures.
In my recent work with Louis Goldstein and Elsi Kaiser (Lee, Goldstein & Kaiser, 2020), using a mouse-tracking method, we examine listeners’ segmentation of internal open juncture sequences, specifically looking at listeners’ sensitivity to sub-segmental information. Our findings suggest that segmentation and lexical access are highly attuned to bottom-up phonetic information, have implications for a model of spoken language recognition with position-specific representations available at the prelexical level, and allude to the possibility that detailed phonetic information may be stored in the listeners’ lexicons.
In my work with Louis Goldstein and Shrikanth Narayanan (Lee, Goldstein & Narayanan, 2015), using rtMRI we shed light on the articulatory composition of the Korean liquid in terms of different prosodic contexts. Our results suggest that the positional allophony between lateral and flap is not only attributable to overall gestural reduction, but also to a categorical distinction in gestural composition.
Speech Accommodation in Dyadic Interaction
A second arm of my research program is the study of the cognitive control of interactional spoken language behavior. Specifically, I examine the adaptive accommodation behaviors in pairs of interacting speakers.
In NIH-funded work on conversational interaction published in PLoS One (Lee, Y., Gordon Danner, Parrell, Lee, S., Goldstein & Byrd, 2018), we employed a dual-EMA setup to investigate how speakers' prosodic, acoustic, and articulatory speech behavior adapts to their dyad partner's speech over the course of an interaction. Our results contribute to an understanding of how the realization of linguistic phrasal structure is coordinated across interacting speakers and provide novel evidence that speakers do exhibit accommodation with one another at the level of cognitively specified motor control of articulatory gestures and that this accommodation is sensitive to prosodic structure.
In our most recent work (Lee, Goldstein, Parrell & Byrd, 2021), we investigate how specific aspects of individual variability may shape the way in which conversing interlocutors adapt their speech behaviors. We construct a simple computational model of attunement and present the model simulation results and behavioral data of the convergence behaviors of two interlocutors conversing over time. Our findings demonstrate the key contribution of individual variability or flexibility to speaker adaptability and identify this structured variability as an important factor in determining who converges in a spoken language interaction exhibiting accommodation.
Voice Identity from Variability
In my current postdoctoral project funded by NIH and NSF, I am collaborating with Jody Kreiman, Patricia Keating, and Abeer Alwan, pursuing an interdisciplinary approach to the study of voice quality. Kreiman and her team have identified a suite of measures that constitute a psychoacoustic model of voice quality, paving the way for a long overdue refinement in characterizing voice variation both within and between speakers. The voice quality field has mainly focused on the variability between speakers, and the nature and importance of everyday variability in voice quality within a speaker have been largely overlooked. I have been tackling the challenge of identifying measures of voice quality that account for perceptually relevant acoustic variance both within and between speakers. With a large voice dataset of within- and between-speaker variability, my work significantly contributes to our understanding of how voice spaces—both individually (within-talker) and generally (between-talkers)—are formulated with reference to acoustic attributes. This investigation further reveals that similarity in these reference patterns plays an important role in voice recognition. Our research informs cross-linguistic accounts of voice quality and serves as a basis for studies on voice production and recognition and for clinical diagnosis and treatment of deviation in voice quality. (Associated Work: Lee & Kreiman, 2019, 2020; Lee, Keating & Kreiman, 2019, Lee & Kreiman, 2019; Kreiman & Lee, 2019, 2020; Lee & Kreiman, 2021; Lee, Garellek, Esposito & Kreiman, 2021)